Assessing Helicopter Turbine Engine Health: A Simple Yet Robust
Probabilistic Approach
Peihua Han
1
, Qin Liang
2
, Erik Vanem
3
, Knut Erik Knutsen
4
, Houxiang Zhang
5
1,2,5
Department of Ocean Operations and Civil Engineering, Norwegian University of Science and Technology,
˚
Alesund, Norway
peihua.han@ntnu.no
qinlia@stud.ntnu.no
hozh@ntnu.no
2,3,4
Group Research and Development - DNV, Høvik, Norway
Qin.Liang@dnv.com
Erik.Vanem@dnv.com
Knut.Erik.Knutsen@dnv.com
ABSTRACT
This paper presents a data-driven approach for assessing the
health of helicopter turbine engines, developed for the PHM
North America 2024 Conference Data Challenge. The task
involves both regression and classification to estimate the
torque margin and classify engine health as either nominal
or faulty. To quantify the reliability of predictions, proba-
bilistic outputs are generated. We employ a two-stage model
where the predicted torque margin serves as an input feature
for health classification. For probabilistic torque margin es-
timation, we introduce an empirical error sampling method
to generate torque margin samples, followed by a rule-based
distribution selection scheme to evaluate the resulting distri-
butions. For fault classification, logistic regression is used
to provide confidence estimates, and we incorporate a score-
optimized loss function during training to mitigate penalties
for false negatives. Our approach demonstrates strong gen-
eralization to unseen assets, ranking 2nd in the competition
with a score of 0.94, demonstrating its effectiveness in pre-
dicting health conditions and uncertainty for more informed
helicopter engine management.
1. INTRODUCTION
A turbine engine, also known as a gas turbine or jet engine, is
an internal combustion engine that turns fuel into mechanical
energy. In helicopters, these engines are crucial for provid-
ing the power needed for lift and maneuvering. They operate
under high-stress conditions and experience significant wear
and tear, which can lead to failures. Effective health assess-
Peihua Han et al. This is an open-access article distributed under the terms of
the Creative Commons Attribution 3.0 United States License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the
original author and source are credited.
ment is essential to predict potential problems, prevent major
failures, and lower maintenance costs. It can ensure that en-
gines operate safely and efficiently, supporting reliable heli-
copter performance.
Prognostics and Health Management (PHM) is an integrated
framework designed to monitor, diagnose, and predict the
condition of systems (Zhang et al., 2022). It has been ap-
plied to many applications including turbine engines. PHM
encompasses three key components: anomaly detection (Han,
Ellefsen, Li, Holmeset, & Zhang, 2021), fault diagnos-
tics (Wang et al., 2020), and fault prognostics (Han, Ellefsen,
Li, Æsøy, & Zhang, 2021). PHM methods can be categorized
into model-based and data-driven approaches depending on
whether a physical model is used. Data-driven methods have
gained popularity in PHM due to their ability to handle com-
plex, high-dimensional data and uncover patterns that may be
challenging for traditional model-based approaches to cap-
ture (Liang, Knutsen, Vanem, Æsøy, & Zhang, 2024).
In data-driven settings, fault diagnostics can be addressed
through regression or classification. When using regression,
the target output typically represents the system’s degrada-
tion level, modeled as a continuous variable. For example,
Vanem et al. (2023); Liang, Vanem, et al. (2023) estimated
the state of health of a battery by extracting features from
charging and discharging curves and applying various sta-
tistical models. Similarly, Mathew et al. (2024) developed
a one-dimensional convolutional neural network to estimate
the capacity factor of wind farms, utilizing the Huber loss
function to mitigate the impact of outliers. In classification
tasks, the target output is a categorical variable, e.g., fault
detection often involves binary outputs, while fault isolation
deals with multiclass outputs that represent specific isolation
1
